diff --git a/docs/research/2026-09-01-overhaul/s4-depth-alpha-packet.md b/docs/research/2026-09-01-overhaul/s4-depth-alpha-packet.md
index f21016de..c2980af4 100644
--- a/docs/research/2026-09-01-overhaul/s4-depth-alpha-packet.md
+++ b/docs/research/2026-09-01-overhaul/s4-depth-alpha-packet.md
@@ -2048,6 +2048,27 @@ Automated return:
   No new failure appeared; evidence
   `artifacts/s4-c3b-installed-inclusive-20260904`.
 
+Artifact provenance: both official artifact sets above ran before the
+implementation commit, on the restored but still uncommitted c3b tree whose
+HEAD was parent `a0523e6057af206126a09874bf8664206ff7d9a1`. Accordingly,
+both embedded summaries truthfully record that parent and
+`WorktreeDirty=true`, rather than implementation tip `4eca2a4f`. The two
+`SHA256SUMS.txt` manifests each verify all 30 entries with zero missing or
+mismatched files.
+
+Both artifact environments list
+`WalkTraceConformanceTests.AlphaDepthTranscript.cs` as dirty. That entry was
+transient mutation-worktree dirtiness left by applying and restoring mutation
+5, not an active prose mutation during either official gate: the hermetic TRX
+records
+`AlphaFlushCountKnownFailure_StatesOnlyTheThreeExactReasons` as `Passed`, and
+the transcript has the identical Git blob
+`c755c1379feeba5fe5382fa0b0db243e7169b4e9` at parent `a0523e605`,
+implementation tip `4eca2a4f`, current contract HEAD, and the current
+worktree. The final implementation diff therefore correctly excludes that
+file while the pre-commit artifacts continue to describe their actual capture
+state.
+
 Mutation ledger (each mutation was applied alone and restored by patch before
 the next; the restored pin set passed 4/4):